A method for tracking small infrared targets using grayscale difference angle enhancement and logarithmic threshold update

Abstract

This invention provides an infrared small target tracking method based on grayscale difference angle enhancement and logarithmic threshold update, relating to the field of infrared small target tracking technology. The method includes: S101, acquiring the current frame's original infrared image; S102, based on the predicted centroid coordinates of the previous frame's original infrared image and the current frame's original infrared image, sequentially performing target region selection and grayscale decrease angle calculation to obtain a final response map; S103, performing quality assessment on the final response map to obtain a quality assessment result; S104, based on the quality assessment result and combined with the final response map or the current frame's original infrared image, predicting the current predicted centroid coordinates; S105, using the current predicted centroid coordinates and the predicted centroid coordinates corresponding to all previous original infrared images before the current frame's original infrared image to form the final tracking result. This achieves robust and continuous tracking of infrared small targets, improving the practicality and stability of engineering applications.

Description

Technical Field

[0001] This invention relates to the field of infrared small target tracking technology, specifically to an infrared small target tracking method that combines grayscale difference angle enhancement and logarithmic threshold update. Background Technology

[0002] Infrared small target tracking is a key technology in public safety and disaster relief, early warning and surveillance, and other fields. Infrared small targets in images typically appear as tiny points or small patches with limited texture information and low contrast to the background. Because infrared imaging is susceptible to interference from complex backgrounds and sensor noise, the features of infrared small targets are easily obscured. Furthermore, infrared small targets may exhibit dynamic characteristics such as rapid maneuvering, scale changes, and brief disappearances, making stable and robust tracking a significant challenge. Especially under complex backgrounds and low signal-to-noise ratio conditions, effectively distinguishing infrared small targets from interference, maintaining model stability under interference, responding to the movement and appearance changes of infrared small targets, and quickly relocating them after loss of tracking are core problems that current technology urgently needs to solve.

[0003] To address the aforementioned challenges, existing technologies have proposed and improved various tracking schemes based on correlation filtering. One typical implementation is the ECO-SWGST (Efficient Convolution Operators – Spatial-Weighted Guided Search Tracker) method. This method employs a two-stage framework of "detection enhancement followed by correlation filtering and tracking." In the detection stage, the method uses the side-window generalized structure tensor to analyze the image, aiming to suppress background edge structures while enhancing the feature response of the infrared small target itself to improve its saliency in subsequent processing. In the tracking stage, the enhanced features are input into the correlation filter to complete the localization and scale estimation of the infrared small target. The filter model is then updated online based on a single quality metric such as the peak-to-sidelobe ratio of the current frame response map to maintain tracking continuity in complex scenes.

[0004] Despite the progress made by existing methods, significant shortcomings remain in practical engineering applications. First, at the feature level, existing methods largely rely on manually designed gradients or features. These features have limited distinguishing ability when infrared small targets are similar in grayscale to the background. The feature enhancement process often indiscriminately strengthens all local structures in the image, potentially amplifying noise or other interference while suppressing some background, making it difficult to generate a sustained, sharp, and unique peak in the response map. Second, regarding filter model update strategies, the simple criteria used in existing methods, based on the quality of a single frame response map, are prone to misjudgment when infrared small targets encounter brief occlusion, rapid movement, or sudden brightness changes. This can lead to the filter model learning incorrect background information, causing filter model contamination and subsequent tracking drift. Finally, when infrared small targets are lost, existing re-detection mechanisms often employ global search or fixed threshold judgment, resulting in low computational efficiency and a high false alarm rate in complex backgrounds. This makes it difficult to achieve fast and accurate relocalization, affecting the overall practicality and continuity of tracking. Summary of the Invention

[0005] To address the aforementioned problems in the prior art, this invention provides an infrared small target tracking method based on grayscale difference angle enhancement and logarithmic threshold update.

[0006] The technical problem to be solved by this invention is achieved through the following technical solution: This invention provides an infrared small target tracking method based on grayscale difference angle enhancement and logarithmic threshold update, comprising: S101. Obtain the original infrared image of the current frame; S102. Based on the predicted centroid coordinates of the previous frame's original infrared image and the current frame's original infrared image, the target region selection and grayscale decrease angle calculation are performed sequentially to obtain the final response map. S103. Perform quality assessment on the final response map to obtain the quality assessment result; S104. Based on the quality judgment result and combined with the final response map or the original infrared image of the current frame, the current predicted centroid coordinates are obtained. S105. Using the current predicted centroid coordinates and the predicted centroid coordinates of all previous original infrared images before the current frame's original infrared image, the final tracking result is formed. The grayscale decrease angle calculation is the arctangent angle value corresponding to the grayscale change rate of each pixel in the neighborhood direction in the selected target area. The quality assessment results include: current ideal filter normal update, current ideal filter slightly contaminated, and current ideal filter severely contaminated. The current ideal filter normal update indicates that the prediction process should continue according to the current ideal filter. The current ideal filter slightly contaminated indicates that an update strategy should be adopted to update the current ideal filter. The current ideal filter severely contaminated indicates that a re-detection process should be performed on the original infrared image of the current frame.

[0007] This invention provides an infrared small target tracking method based on grayscale difference angle enhancement and logarithmic threshold update, comprising: S101, acquiring the current frame's original infrared image; S102, based on the predicted centroid coordinates of the previous frame's original infrared image and the current frame's original infrared image, sequentially performing target region selection and grayscale decrease angle calculation to obtain a final response map; S103, performing quality judgment on the final response map to obtain a quality judgment result; S104, based on the quality judgment result and combined with the final response map or the current frame's original infrared image, predicting the current predicted centroid coordinates; S105, using the current predicted centroid coordinates and all previous original infrared images of the current frame's original infrared image... The predicted centroid coordinates corresponding to the external image form the final tracking result; the gray-level descent angle calculation is the arctangent angle value corresponding to the gray-level change rate of each pixel in the selected target region in the neighborhood direction; the quality judgment results include: current ideal filter normal update, current ideal filter slightly contaminated, and current ideal filter severely contaminated; current ideal filter normal update is used to indicate that the prediction process should continue according to the current ideal filter; current ideal filter slightly contaminated is used to indicate that an update strategy should be adopted to update the current ideal filter; current ideal filter severely contaminated is used to indicate that a re-detection process should be performed on the original infrared image of the current frame. In this invention, firstly, the gray-level descent angle calculation is used as a feature enhancement method to generate robust features based on the arctangent angle value of the gray-level change rate of the pixel neighborhood, which solves the problem that artificial features have limited distinguishing ability and are prone to amplifying noise or interference when the gray levels of the target and the background are similar, thereby generating a continuous, sharp, and unique peak in the response image. Then, this invention designs a three-level intelligent quality judgment and current ideal filter update strategy, solving the problems of misjudgment, model contamination, and tracking drift caused by traditional single-frame simple criteria when the target encounters occlusion or sudden changes. This achieves safer and more controllable management of the current ideal filter update process. Finally, when the quality judgment result is "current ideal filter severely contaminated," this invention triggers a re-detection process for the original infrared image of the current frame, replacing the inefficient global search or fixed threshold method with a high false alarm rate. This solves the problems of high computational cost and poor accuracy of relocation after target loss, ensuring the continuity and practicality of tracking. In summary, this invention, through the synergy of feature enhancement, intelligent updating, and efficient re-detection, achieves robust and continuous tracking of small infrared targets, improving the practicality and stability of engineering applications.

[0008] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0009] Figure 1 A schematic flowchart of an infrared small target tracking method with gray-scale difference angle enhancement and logarithmic threshold update provided in an embodiment of the present invention; Figure 2 An illustrative diagram shows the gray-gradient consistency scores of different candidate regions obtained based on the method of the present invention. Detailed Implementation

[0010] The present invention will be further described in detail below with reference to specific embodiments, but the implementation of the present invention is not limited thereto.

[0011] To achieve robust and continuous tracking of small infrared targets, this invention provides a method for tracking small infrared targets by enhancing grayscale difference angle and updating logarithmic threshold. Figure 1 This is a flowchart illustrating an infrared small target tracking method based on grayscale difference angle enhancement and logarithmic threshold update, as provided in an embodiment of the present invention. Figure 1 As shown, it includes: S101. Obtain the original infrared image of the current frame.

[0012] It should be noted that in a fire rescue scenario, the current frame of the raw infrared image can be a raw thermal radiation data frame directly captured by an infrared thermal imager without subsequent algorithm processing. The final output tracking result can be a trajectory formed by the real-time location coordinates of the trapped person in the fire.

[0013] S102. Based on the predicted centroid coordinates of the previous frame's original infrared image and the current frame's original infrared image, the target region selection and grayscale decrease angle calculation are performed sequentially to obtain the final response map.

[0014] Among them, the grayscale decrease angle calculation is the arctangent angle value corresponding to the grayscale change rate of each pixel in the neighborhood direction in the result of the selected target area calculation.

[0015] Optionally, S102 includes: Using the predicted centroid coordinates of the previous frame's original infrared image, the target region is selected at the corresponding position in the current frame's original infrared image, forming a slice of the current frame. Calculate the grayscale decrease angle on the current frame slice to obtain grayscale decrease information; The pixel update process of the current frame slice image is performed using the grayscale decrease information to obtain the final response image.

[0016] Optionally, the grayscale decrease information is represented as: ; in, express Direction first Grayscale decrease information between each pixel and the current pixel. Indicates the first One direction, , Indicates the first 1 pixel , Indicates along direction The total number of pixels examined express Current pixel in the direction and the first Gray-scale difference of individual pixels express Current pixel in the direction and the first Euclidean distance between pixels Represents the protective constant. This indicates the calculation of the arctangent angle.

[0017] It should be noted that in this invention, the slice map of the initial frame original infrared image can be obtained in two ways: one is manual pre-selection; the other is automatic selection using a target detection algorithm. Specifically, after obtaining the initial frame original infrared image, the approximate range of the target can be manually marked, or it can be input into a pre-trained target detection model for selection. The resulting area is used as the slice map of the initial frame original infrared image.

[0018] Optionally, the pixel update process of the current frame slice image is performed using the grayscale decrement information to obtain the final response image, including: Obtain the maximum value of grayscale decrease information for each pixel in the current frame slice image across all directions, and obtain multiple maximum grayscale decrease information; The average value of each pixel is obtained by averaging the multiple maximum grayscale decrease information corresponding to each pixel in the current frame slice image. Replace the pixel values ​​at the corresponding pixel points in the current frame slice with saliency values ​​to obtain the initial response map; Gaussian filtering is applied to the initial response map to obtain a multi-scale enhanced feature map. The initial response map and the multi-scale enhanced feature map are weighted and fused to obtain the final response map.

[0019] Optionally, the initial response diagram is represented as: ; in, Indicates the current pixel. The significance value corresponding to the initial response plot, The x-axis represents the initial response plot. The ordinate of the initial response plot is represented by the vertical axis. This indicates the number of valid directions out of all directions. Indicates the first One direction, , express Direction first Grayscale decrease information between each pixel and the current pixel. Indicates the first 1 pixel , Indicates along direction The total number of pixels examined.

[0020] In addition, the Gaussian filtering process in the embodiments of the present invention can also be replaced by other multi-scale methods, such as processing flow based on Laplace pyramid or wavelet multi-resolution decomposition.

[0021] S103. Perform quality assessment on the final response map to obtain the quality assessment result.

[0022] In this invention, after obtaining the final response map, its quality needs to be assessed to determine whether the prediction process for the current frame was successful. Therefore, this invention constructs a log-weighted confidence score based on the final response map, and obtains the corresponding quality assessment result by comparing the log-weighted confidence score with a first threshold and a second threshold. The quality assessment results include: the current ideal filter is updated normally, the current ideal filter is slightly contaminated, and the current ideal filter is severely contaminated.

[0023] Specifically, S103 includes: Calculate the log-weighted confidence level corresponding to the final response map; The log-weighted confidence score is compared with the first and second thresholds. If the log-weighted confidence score is greater than the first threshold, the quality judgment result is that the current ideal filter is updated normally. If the log-weighted confidence score is less than the second threshold, the quality assessment result is that the current ideal filter is severely contaminated. If the log-weighted confidence level is less than or equal to the first threshold and greater than or equal to the second threshold, the quality assessment result is that the current ideal filter is slightly contaminated.

[0024] Optionally, the log-weighted confidence level is expressed as: ; in, This represents the log-weighted confidence level. Indicates the weighting coefficient. This represents the maximum pixel value in the final response image. This represents the average maximum pixel value corresponding to the final response map of the historical frame. This represents the average peak correlation energy of the final response plot. This represents the mean peak correlation energy of the final response map of historical frames.

[0025] Furthermore, in some possible implementations, the calculation of logarithmically weighted confidence scores is not limited to the logarithmic function form mentioned above; other forms such as exponential functions, power functions, or sigmoid functions can also be used. As long as the constructed function can effectively fuse the two key types of information—peak features and pixel value distribution features—the same confidence score discrimination effect can be achieved.

[0026] S104. Based on the quality judgment result and combined with the final response map or the original infrared image of the current frame, the current predicted centroid coordinates are obtained.

[0027] In this invention, when the quality judgment result is that the current ideal filter is updated normally, the prediction process is continued according to the current ideal filter; when the quality judgment result is that the previous ideal filter is slightly contaminated, the update strategy is adopted to update the current ideal filter; when the quality judgment result is that the previous ideal filter is severely contaminated, the re-detection process is performed on the original infrared image of the current frame.

[0028] Optionally, when the quality judgment result indicates that the current ideal filter has been updated normally, S104 includes: The current predicted centroid coordinates are obtained by performing relevant calculations on the final response map using the current ideal filter.

[0029] Optionally, when the quality determination result is that the current ideal filter is slightly contaminated, S104 includes: An update strategy is adopted to update the current ideal filter, resulting in an updated ideal filter; The updated ideal filter is used to perform relevant calculations on the final response map to obtain the current predicted centroid coordinates. The update strategy is expressed as: ; in, Indicates the first The updated ideal filter corresponding to the original infrared image of the frame. , This indicates the total number of frames in the original infrared image. Indicates the update coefficients. Indicates the first The front of the original infrared image An ideal filter is preserved in the original infrared image frame. , Indicates the first The current ideal filter corresponding to the original infrared image frame. This indicates a scalar multiplication operation.

[0030] ; in, Denotes the base of the natural logarithm. This represents the total number of frames in the original infrared image of the historical frames with a log-weighted confidence level less than 0.4. This represents the attenuation coefficient.

[0031] It should be noted that in some other possible implementations, a long-short-term strategy constrained by log-weighted confidence can be used to update the current ideal filter. Specifically, the strength of the update strategy can be dynamically adjusted according to the magnitude of the log-weighted confidence; that is, the larger the log-weighted confidence, the larger the update coefficient can be set.

[0032] Optionally, when the quality judgment result indicates that the current ideal filter is severely contaminated, S104 includes: The original infrared image of the current frame is divided into regions with all the original infrared images preceding the original infrared image of the current frame to obtain multiple image blocks; Multiple image blocks are stitched together in chronological order to obtain the image stitching tensor; The Jacobi loop method is used to decorrelate the image stitching tensor to obtain a decorrelated image. Gaussian filtering is applied to the decorrelation image to obtain the decorrelation-enhanced feature map; The density peak search method is used to filter candidate points in the decorrelation-enhanced feature map to obtain information on multiple candidate points. The seed region growing method is used to process multiple candidate point information for candidate point judgment and obtain the final re-detection result; The authenticity of the target features is determined based on the final re-detection results to obtain the authenticity determination result; Based on the authenticity determination result, and combined with the final re-detection result or the predicted centroid coordinates of the previous frame of the original infrared image, the current predicted centroid coordinates are obtained. Among them, the determination of the authenticity of target features is to determine the brightness concentration and edge direction consistency of the final re-detection result.

[0033] In this embodiment of the invention, the authenticity of the target features in the final redetection result is measured based on the gray-gradient consistency score.

[0034] The gray-gradient consistency score is represented as: ; ; ; in, This represents the gray-level gradient consistency score. This represents the product metric indicating the decrease in grayscale value of the final redetection result. The local gradient consistency factor represents the final re-detection result. Indicates the first That's the right direction. , This represents the brightness value of the central bright spot in the final re-detection result. Indicates along the direction The first direction in The local minimum gray value obtained by the search Indicates along the direction The second direction The local minimum gray value obtained by the search This indicates the consistency of gradient directions in the final re-detection results. This represents the mean of the gradient magnitudes in the final re-detection results. Indicates the direction is correct The first direction in the middle, Indicates the direction is correct The second direction.

[0035] Figure 2 An exemplary diagram illustrates the gray-gradient consistency scores of different candidate points obtained based on the method of the present invention. Figure 2 The core objective is to determine whether a candidate point is a real target by quantitatively evaluating its "brightness concentration" and "edge direction consistency." For example... Figure 2 As shown in the image, against a typical infrared scene background, two candidate points for comparison are highlighted with yellow and red boxes: the candidate point marked with the yellow box... Low value ( The value is only 9.42), which intuitively indicates that the candidate point may be noise or interference; while the red box marks a more significant potential target, whose outline is clearer in the accompanying enlarged inset and shows a higher [value / indicator]. value( The value is 116.34, which comprehensively reflects the candidate point's smooth gray-level decrease and strong gradient direction consistency in multiple directions, thus possessing true target characteristics. The blurred texture in the background further highlights the salience of the candidate point. The overall composition, through numerical annotation and visual comparison, intuitively verifies... The value is used to distinguish the effectiveness of true and false targets.

[0036] In other implementations, the gray-gradient consistency score can be replaced by a structural discriminant measure such as eigenvalue ratio, energy concentration, or principal direction consistency, as long as it can reflect the structural saliency and target consistency of the candidate region corresponding to the predicted centroid coordinates.

[0037] It should be noted that the authenticity determination result is either successful re-detection or failed re-detection.

[0038] After obtaining the gray-level-gradient consistency score, it is compared with a preset ground truth threshold. When the gray-level-gradient consistency score is greater than or equal to the ground truth threshold, it indicates that the redetection is successful, and the final redetection result can be used as the current predicted centroid coordinates. When the gray-level-gradient consistency score is less than the ground truth threshold, it indicates that the redetection has failed, and the predicted centroid coordinates of the previous frame's original infrared image can be used as the current predicted centroid coordinates.

[0039] It should be noted that in some other possible implementations, when the quality assessment result indicates that the current ideal filter is severely contaminated, the current predicted centroid coordinates can be determined based on historical frame consistency constraints without initiating a re-detection process. This approach not only improves detection efficiency but also enhances the system's ability to suppress noise, clutter, and false targets.

[0040] S105. Using the current predicted centroid coordinates and the predicted centroid coordinates of all previous original infrared images before the current frame's original infrared image, the final tracking result is formed.

[0041] In this embodiment, the predicted centroid coordinates of the current frame and the predicted centroid coordinates of all previous original infrared images are stitched together in chronological order to form the final tracking result.

[0042] This invention provides an infrared small target tracking method based on gray-level difference angle enhancement and logarithmic threshold update. First, a feature enhancement method is used, employing gray-level decrease angle calculation. Robust features are generated based on the arctangent angle of the gray-level change rate in the pixel neighborhood. This addresses the limitations of manual features in distinguishing between the target and background when their gray levels are similar, and their tendency to amplify noise or interference, resulting in a continuous, sharp, and unique peak in the response image. Second, this invention designs a three-level intelligent quality judgment and current ideal filter update strategy. This solves the problems of traditional single-frame simple criteria leading to misjudgment, model contamination, and tracking drift when the target encounters occlusion or abrupt changes. It achieves safer and more controllable management of the current ideal filter update process. Finally, when the quality judgment result is "current ideal filter severely contaminated," this invention triggers a re-detection process for the original infrared image of the current frame. This replaces the inefficient and high-false-alarm-rate global search or fixed-threshold methods, solving the problems of high computational cost and poor accuracy in relocation after target loss, ensuring the continuity and practicality of tracking. In summary, this invention achieves robust and continuous tracking of small infrared targets through the synergy of feature enhancement, intelligent updating, and efficient re-detection, thereby improving the practicality and stability of engineering applications.

[0043] To verify the effectiveness of the method of the present invention, in this embodiment, the method of the present invention was compared with nine typical infrared small target tracking methods on a public dataset of infrared small targets, based on two core indicators: DP (Distance Precision) and OP (Overlap Precision). The comparison results are shown in Table 1. Nine typical infrared small target tracking methods used for comparison are: ARCF (Aberrance Repressed Correlation Filters), Auto-Tracker (Automatic Spatio-Temporal Regularized Correlation Filters), BACF (Background-Aware Correlation Filters), ESFCF (Edge and Semantic Feature Correlation Filter), STRCF (Spatial-Temporal Regularized Correlation Filters), ECO-SWGST (ECO-based tracking with Side Window Generalized Structure Tensor), Tomp (Transforming Model Prediction for Tracking), Siam-IST (Infrared Small Target Tracking based on an improved SiamRPN), LADCF (Learning... Adaptive Discriminative Correlation Filters (learning adaptive discriminative correlation filtering algorithms). These comparative methods cover a variety of technical approaches, from classic correlation filtering frameworks to emerging deep feature enhancement methods and structure tensor-based feature enhancement methods, reflecting the capabilities of the proposed method in terms of feature representation, anti-interference performance, and stability in complex backgrounds from different dimensions.

[0044] Table 1 Performance Comparison of Different Tracking Methods

[0045] As shown in Table 1, the method of this invention achieved the highest scores in both core metrics (DP: 77.36%, OP: 74.20%), significantly outperforming all compared algorithms. This indicates that the method of this invention has a greater advantage in infrared small target localization and tracking. Analysis reveals that this advantage mainly stems from two core mechanisms introduced in the method of this invention: First, by calculating the grayscale descent angle, the arctangent angle value of the grayscale change rate of each pixel in the neighborhood direction is extracted, effectively highlighting the difference between small targets and complex backgrounds; Second, by introducing a quality judgment mechanism for the final response map, and determining whether to continue tracking, update the filter, or initiate re-detection based on the quality judgment result, an adaptive and robust filter maintenance strategy is implemented. The combination of these two aspects enables the tracking system to maintain the sharp peak of the response map in complex backgrounds, thereby achieving higher localization and tracking accuracy.

[0046] The method provided in this embodiment of the invention can be applied to electronic devices. Specifically, the electronic device can be a desktop computer, a portable computer, a smart mobile terminal, a server, etc., and this embodiment of the invention does not limit the application to such devices.

[0047] Although the invention has been described herein in conjunction with various embodiments, those skilled in the art, by reviewing the accompanying drawings and the disclosure, will understand and implement other variations of the disclosed embodiments in carrying out the claimed invention. In this description, the word "comprising" does not exclude other components or steps, "a" or "an" does not exclude a plurality, and "a plurality" means two or more, unless otherwise explicitly specified. Furthermore, while different embodiments may describe certain measures, this does not mean that these measures cannot be combined to produce good results.

[0048] The above description, in conjunction with specific preferred embodiments, provides a further detailed explanation of the present invention. It should not be construed that the specific implementation of the present invention is limited to these descriptions. For those skilled in the art, various simple deductions or substitutions can be made without departing from the inventive concept, and all such modifications and substitutions should be considered within the scope of protection of the present invention.

Claims

1. A method for tracking small infrared targets using grayscale difference angle enhancement and logarithmic threshold update, characterized in that, include: S101. Obtain the original infrared image of the current frame; S102. Based on the predicted centroid coordinates of the previous frame's original infrared image and the current frame's original infrared image, the target region selection and grayscale decrease angle calculation are performed sequentially to obtain the final response image. S103. Perform a quality assessment on the final response map to obtain a quality assessment result; S104. Based on the quality judgment result and combined with the final response map or the original infrared image of the current frame, the current predicted centroid coordinates are obtained. S105. Using the current predicted centroid coordinates and the predicted centroid coordinates corresponding to all previous original infrared images before the current frame's original infrared image, a final tracking result is formed. The grayscale decrease angle calculation is the arctangent angle value corresponding to the grayscale change rate of each pixel in the neighborhood direction in the result of the selected target region; The quality assessment results include: the current ideal filter is updated normally, the current ideal filter is slightly polluted, and the current ideal filter is severely polluted; The "normal update of the current ideal filter" indicates that the prediction process should continue according to the current ideal filter; the "slightly polluted current ideal filter" indicates that an update strategy should be adopted to update the current ideal filter; and the "severely polluted current ideal filter" indicates that a re-detection process should be performed on the original infrared image of the current frame.

2. The infrared small target tracking method according to claim 1, characterized in that, S102 includes: Using the predicted centroid coordinates of the previous frame of the original infrared image, the target region is selected at the corresponding position in the current frame of the original infrared image to form a slice of the current frame. The grayscale decrease angle is calculated on the current frame slice to obtain grayscale decrease information; The grayscale decrease information is used to update the pixels of the current frame slice to obtain the final response image.

3. The infrared small target tracking method according to claim 2, characterized in that, The grayscale decrease information is represented as follows: ； in, express Direction first Grayscale decrease information between each pixel and the current pixel. Indicates the first One direction, , Indicates the first 1 pixel , Indicates along direction The total number of pixels examined express Current pixel in the direction and the first Gray-scale difference of individual pixels express Current pixel in the direction and the first Euclidean distance between pixels Represents the protective constant. This indicates the calculation of the arctangent angle.

4. The infrared small target tracking method with gray-scale difference angle enhancement and logarithmic threshold update according to claim 3, characterized in that, The step of updating the pixel image of the current frame slice using the grayscale decrement information to obtain the final response image includes: Obtain the maximum value of the grayscale decrease information corresponding to each pixel in all directions in the current frame slice image to obtain multiple maximum grayscale decrease information; The average value of each pixel is obtained by averaging the multiple maximum grayscale decrease information corresponding to each pixel in the current frame slice image. The pixel values ​​at the corresponding pixel points in the current frame slice are replaced with the saliency values ​​to obtain the initial response map; The initial response map is subjected to Gaussian filtering to obtain a multi-scale enhanced feature map; The initial response map and the multi-scale enhanced feature map are weighted and fused to obtain the final response map.

5. The infrared small target tracking method according to claim 4, characterized in that, The initial response diagram is represented as follows: ； in, Indicates the current pixel. The significance value corresponding to the initial response map, The x-axis represents the initial response graph. The ordinate of the initial response graph is represented by the vertical axis. This indicates the number of valid directions out of all directions. Indicates the first One direction, , express Direction first Grayscale decrease information between each pixel and the current pixel. Indicates the first 1 pixel , Indicates along direction The total number of pixels examined.

6. The infrared small target tracking method according to claim 1, characterized in that, S103 includes: Calculate the log-weighted confidence level corresponding to the final response map; The log-weighted confidence score is compared with the first threshold and the second threshold. When the log-weighted confidence level is greater than the first threshold, the quality judgment result is that the current ideal filter is updated normally. If the log-weighted confidence level is less than the second threshold, the quality judgment result is that the current ideal filter is severely contaminated. If the log-weighted confidence level is less than or equal to the first threshold and greater than or equal to the second threshold, then the quality determination result is that the current ideal filter is slightly contaminated.

7. The infrared small target tracking method according to claim 6, characterized in that, The log-weighted confidence level is expressed as follows: ； in, This represents the log-weighted confidence level. Indicates the weighting coefficient. This represents the maximum pixel value in the final response image. This represents the average maximum pixel value corresponding to the final response map of the historical frame. This represents the average peak correlation energy of the final response map. This represents the mean peak correlation energy of the final response map of historical frames.

8. The infrared small target tracking method according to claim 1, characterized in that, When the quality determination result indicates that the current ideal filter has been updated normally, S104 includes: The current ideal filter is used to perform correlation calculations on the final response map to obtain the current predicted centroid coordinates.

9. The infrared small target tracking method according to claim 1, characterized in that, When the quality determination result is that the current ideal filter is slightly polluted, S104 includes: An update strategy is adopted to update the current ideal filter, resulting in an updated ideal filter; The updated ideal filter is used to perform correlation calculations on the final response map to obtain the current predicted centroid coordinates. The update strategy is expressed as follows: ； in, Indicates the first The updated ideal filter corresponding to the original infrared image of the frame. , This indicates the total number of frames in the original infrared image. Indicates the update coefficients. Indicates the first The front of the original infrared image An ideal filter is preserved in the original infrared image frame. , Indicates the first The current ideal filter corresponding to the original infrared image of the frame. This indicates a scalar multiplication operation.

10. The infrared small target tracking method according to claim 1, characterized in that, When the quality judgment result indicates that the current ideal filter is severely contaminated, S104 includes: The original infrared image of the current frame is divided into regions with all the original infrared images preceding the original infrared image of the current frame to obtain multiple image blocks; Multiple image blocks are stitched together in chronological order to obtain an image stitching tensor; The Jacobi cyclic method is used to decorrelate the image stitching tensor to obtain a decorrelated image. The decorrelated image is subjected to Gaussian filtering to obtain a decorrelated enhanced feature map; The density peak search method is used to filter candidate points in the decorrelation-enhanced feature map to obtain information on multiple candidate points. The seed region growing method is used to process the candidate point information of multiple candidate points to obtain the final re-detection result. The final re-detection result is used to determine the authenticity of the target features, and the authenticity determination result is obtained. Based on the authenticity determination result, and combined with the final re-detection result or the predicted centroid coordinates of the previous frame of original infrared image, the current predicted centroid coordinates are obtained. The determination of the authenticity of the target features is a determination of the brightness concentration and edge direction consistency of the final re-detection result.

Images